BYTEC Guides · Article 8 of 8

Introducing AI in your company: four stages, five common mistakes

Introducing AI in your company does not need a big project, it needs a sequence. Most companies now have access to AI, but few have changed anything. Bitkom president Ralf Wintergerst puts it this way: buying a licence means you have bought AI, not that you have changed anything.

AS OF OCTOBER 2026 · 5 MIN. READ

Where companies stand with AI today

In Switzerland, 32% of the companies surveyed use AI in individual areas, 14% across several areas, and 9% have changed their business model with it. 31% are in pilot projects, and 14% have not started anything yet (EY, May 2026). In Germany, 59% of the companies surveyed say they do not use the potential of AI at all, and a further 28% say they use it little (Bitkom, September 2026).

The path from licence to a changed workflow can be taken in four stages. Each is small enough to manage alongside day-to-day business.

Stage 1: an AI usage policy for employees

In many companies, employees have long been using AI, just without rules. In Switzerland, 29% of respondents report that private accounts such as ChatGPT are allowed in their company, and for 8% this is their only access to AI (EY 2026). Only a third of the SMEs that use AI have a clear data protection rule for working with AI, and among companies with 5 to 9 employees it is 23% (AXA, October 2025).

A one-page policy is enough to start with:

Which tools are allowed, with a company account instead of a private one.
Which data must not be entered as long as there is no contract with the provider: customers' personal data, health data, trade secrets.
That a person checks before anything goes out.
Who answers questions internally.

What data protection law requires is set out in AI and data protection in Switzerland.

Stage 2: AI training for employees

Lack of knowledge is the most common reason against AI. In the EU it is cited by 71% of companies that considered using AI and then did not introduce it (Eurostat, 2025 data). In Germany, 85% of companies without AI say they lack the technical know-how. One company in four offers no AI training at all, compared with 43% the year before (Bitkom 2026). In Switzerland, 58% of respondents name expanding AI education and training as the most important measure for the country as a business location (EY 2026).

Good training is concrete: based on your own workflows, with the permitted tools, and with examples of what works well and what does not.

Stage 3: one workflow, one pilot, one measurement

Instead of a little AI everywhere: one workflow that occurs often, measured beforehand, accompanied for about 30 days and then decided on. How this works is described in The first AI pilot. Whether your workflow is suitable is shown by the checklist.

The hurdles are rarely technical. In Switzerland, respondents most often name data quality and data silos (20%), security and data protection concerns (19%) and a lack of skilled staff (18%) as the biggest obstacle (EY 2026).

Stage 4: redesign workflows, then expand

Only when the pilot holds up in measurement is expansion worthwhile. What matters is less the tool than the workflow: of the companies that see a significant financial effect, nearly three quarters have fundamentally redesigned their workflows because of AI, compared with one quarter of the others (McKinsey, August 2026).

The costs also lie elsewhere than expected. German AI users name infrastructure (51%), data preparation (50%) and integration with existing systems (41%) as major cost items. Subscriptions and licences are a major item for only 21% (Bitkom 2026). More on this in What an AI agent really costs.

An AI strategy for an SME fits on one page

Which three workflows come first, who is responsible and how success is measured. How we take this path with companies, from the audit to ongoing operations, is shown in our AI transformation approach.

Five common mistakes in AI projects

Licence instead of workflow. Tools are handed out, but no workflow changes. This is exactly what distinguishes the few successful companies from the rest (McKinsey 2026).
Perception instead of measurement. In the METR trial, experienced developers believed they were 20% faster with AI; measured, they were 19% slower (METR 2025). Without a baseline there is no result.
Shadow AI without rules. Private accounts and no rules for data are everyday reality in many SMEs (see stage 1).
Underestimating data quality. In Switzerland it is the biggest hurdle, in Germany one of the largest cost blocks.
Not bringing employees along. 42% of German AI users name a lack of acceptance among staff as an obstacle (Bitkom 2026).

Frequently asked questions

How long does it take to introduce AI in a company?

There is no general duration; it depends on the workflow. The first pilot is easy to plan: it can be set up for about 30 days of measurement. After that you decide whether and how to expand.

Does an SME need its own AI strategy?

Yes, but a short one: which three workflows come first, who is responsible and how success is measured. Plus a policy on which tools and data are allowed.

Does the EU AI Act also apply to Swiss SMEs?

It can, for example if you offer AI systems in the EU or the output is used there. In Switzerland, the Data Protection Act applies in any case. Our article AI and data protection in Switzerland gives an overview.

Sources

All figures checked against the original sources on 23 October 2026.

Would you like to tackle the first stage in your company? In an initial conversation we identify with you the three workflows that cost the most time, and the most sensible starting point.

Book an initial conversation